DavidAU/Qwen3.5-27B-Claude-4.6-OS-Auto-Variable-Heretic-Uncensored-Thinking

Model Overview

This model, DavidAU/Qwen3.5-27B-Claude-4.6-OS-Auto-Variable-Heretic-Uncensored-Thinking, is a 27 billion parameter variant of the Qwen 3.5 architecture. It has been fine-tuned by DavidAU using four distinct Claude datasets, resulting in improved reasoning capabilities and enhanced output generation compared to the original Qwen 3.5 model. A key differentiator is its "Heretic" training, meaning it is designed to be fully uncensored and highly responsive to user commands without refusal.

Key Capabilities

• Enhanced Reasoning & Output: Fine-tuning on multiple Claude datasets has improved the model's ability to reason and generate high-quality text.
• Uncensored & Compliant: Trained post-"Heretic'ing," the model is uncensored and follows user instructions without safety alignment refusals (14/100 refusals compared to 94/100 for the original Qwen3.5-27B).
• Multimodal (Vision): Vision capabilities (image input) have been tested and confirmed to be working with the new training.
• Improved Tool Handling: Features an upgraded Jinja template to address issues like repetitions and thinking loops, and enhances tool handling.
• Strong Benchmarks: Demonstrates superior performance over the base Qwen3.5-27B-Text-VL model across various benchmarks, including ARC, BoolQ, HSwag, OBQA, PIQA, and Wino.

Good For

• Applications requiring an uncensored and highly compliant language model.
• Tasks benefiting from enhanced reasoning and diverse output generation.
• Multimodal applications involving image understanding.
• Use cases where robust tool handling and reduced generative issues (repetitions, thinking loops) are critical.